GoogleCloudPlatform/kubectl-ai

evals: we should ensure resources in eval tasks must not contain hints to help with the eval tasks

Aperta

#298 aperta il 2 giu 2025

 (0 commenti) (0 reazioni) (0 assegnatari)Go (708 fork)auto 404
enhancementhelp wanted

Metriche repository

Star
 (7532 stelle)
Metriche merge PR
 (Metriche PR in attesa)

Descrizione

We want to ensure the eval tasks represent realistic scenario and we have observed that name/namespaces of the k8s resources used in the task contains hint to make AI Models aware that these are part of evals or some sort of test. I think this affects the model's performance on the task, so we need to fix it for all the existing tasks.

Guida contributor